Question

a) How does PAM (K-medoids) form clusters; how does DBSCAN form clusters?

b) Assume you apply DBSCAN to the same dataset, but the examples in the dataset are sorted differently. Will DBSCAN always return the same clustering for different orderings of the same dataset? Give reasons for your answer.

Answer 1

aANS:          
DBSCAN forms clusrers in the following way:
1.First DBSCAN takes a un-processed core-point 'p' & then forms cluster for this core point which contains all core points & border-points those are density rreachable from point 'p'.
2. This will continue till all the core points has been assigned to the clusters.
   By recursivly computing the points in radius of the core-point leads to form clusters.

k-medoids forms clusters in the simple way below.
k-medoids from clusters by assigning the objects to dataset to the very closest-representative.
Here they use the objects in dataset as the representatves.
.
bANS: The answer is no. The DBSCAN will not return the same clustrng for diff ordernings of simialar.same dataset.
    Reason:
If the dataset containing border-points which are within the radius of 2 or more than 2 core-points, completely different process orders of those core-points can result in different clustering, not same.